// First thing is first. We only want to think about integer here, so if
// we have something in FP form, recast it as integer.
- if (DstEltTy->isFloatingPoint()) {
+ if (DstEltTy->isFloatingPointTy()) {
// Fold to an vector of integers with same size as our FP type.
unsigned FPWidth = DstEltTy->getPrimitiveSizeInBits();
const Type *DestIVTy =
// Okay, we know the destination is integer, if the input is FP, convert
// it to integer first.
- if (SrcEltTy->isFloatingPoint()) {
+ if (SrcEltTy->isFloatingPointTy()) {
unsigned FPWidth = SrcEltTy->getPrimitiveSizeInBits();
const Type *SrcIVTy =
VectorType::get(IntegerType::get(C->getContext(), FPWidth), NumSrcElt);
i != e; ++i, ++GTI) {
ConstantInt *CI = dyn_cast<ConstantInt>(*i);
if (!CI) return false; // Index isn't a simple constant?
- if (CI->getZExtValue() == 0) continue; // Not adding anything.
+ if (CI->isZero()) continue; // Not adding anything.
if (const StructType *ST = dyn_cast<StructType>(*GTI)) {
// N = N + Offset
MapTy = Type::getInt32PtrTy(C->getContext());
else if (LoadTy->isDoubleTy())
MapTy = Type::getInt64PtrTy(C->getContext());
- else if (isa<VectorType>(LoadTy)) {
+ else if (LoadTy->isVectorTy()) {
MapTy = IntegerType::get(C->getContext(),
TD.getTypeAllocSizeInBits(LoadTy));
MapTy = PointerType::getUnqual(MapTy);
APInt ResultVal = APInt(IntType->getBitWidth(), RawBytes[BytesLoaded-1]);
for (unsigned i = 1; i != BytesLoaded; ++i) {
ResultVal <<= 8;
- ResultVal |= APInt(IntType->getBitWidth(), RawBytes[BytesLoaded-1-i]);
+ ResultVal |= RawBytes[BytesLoaded-1-i];
}
return ConstantInt::get(IntType->getContext(), ResultVal);
unsigned StrLen = Str.length();
const Type *Ty = cast<PointerType>(CE->getType())->getElementType();
unsigned NumBits = Ty->getPrimitiveSizeInBits();
- // Replace LI with immediate integer store.
- if ((NumBits >> 3) == StrLen + 1) {
+ // Replace load with immediate integer if the result is an integer or fp
+ // value.
+ if ((NumBits >> 3) == StrLen + 1 && (NumBits & 7) == 0 &&
+ (isa<IntegerType>(Ty) || Ty->isFloatingPointTy())) {
APInt StrVal(NumBits, 0);
APInt SingleChar(NumBits, 0);
if (TD->isLittleEndian()) {
SingleChar = 0;
StrVal = (StrVal << 8) | SingleChar;
}
- return ConstantInt::get(CE->getContext(), StrVal);
+
+ Constant *Res = ConstantInt::get(CE->getContext(), StrVal);
+ if (Ty->isFloatingPointTy())
+ Res = ConstantExpr::getBitCast(Res, Ty);
+ return Res;
}
}
unsigned BitWidth =
TD->getTypeSizeInBits(TD->getIntPtrType(Ptr->getContext()));
- APInt BasePtr(BitWidth, 0);
- bool BaseIsInt = true;
- if (!Ptr->isNullValue()) {
- // If this is a inttoptr from a constant int, we can fold this as the base,
- // otherwise we can't.
- if (ConstantExpr *CE = dyn_cast<ConstantExpr>(Ptr))
- if (CE->getOpcode() == Instruction::IntToPtr)
- if (ConstantInt *Base = dyn_cast<ConstantInt>(CE->getOperand(0))) {
- BasePtr = Base->getValue();
- BasePtr.zextOrTrunc(BitWidth);
- }
-
- if (BasePtr == 0)
- BaseIsInt = false;
- }
// If this is a constant expr gep that is effectively computing an
// "offsetof", fold it into 'cast int Size to T*' instead of 'gep 0, 0, 12'
APInt Offset = APInt(BitWidth,
TD->getIndexedOffset(Ptr->getType(),
(Value**)Ops+1, NumOps-1));
+ Ptr = cast<Constant>(Ptr->stripPointerCasts());
+
+ // If this is a GEP of a GEP, fold it all into a single GEP.
+ while (GEPOperator *GEP = dyn_cast<GEPOperator>(Ptr)) {
+ SmallVector<Value *, 4> NestedOps(GEP->op_begin()+1, GEP->op_end());
+
+ // Do not try the incorporate the sub-GEP if some index is not a number.
+ bool AllConstantInt = true;
+ for (unsigned i = 0, e = NestedOps.size(); i != e; ++i)
+ if (!isa<ConstantInt>(NestedOps[i])) {
+ AllConstantInt = false;
+ break;
+ }
+ if (!AllConstantInt)
+ break;
+
+ Ptr = cast<Constant>(GEP->getOperand(0));
+ Offset += APInt(BitWidth,
+ TD->getIndexedOffset(Ptr->getType(),
+ (Value**)NestedOps.data(),
+ NestedOps.size()));
+ Ptr = cast<Constant>(Ptr->stripPointerCasts());
+ }
+
// If the base value for this address is a literal integer value, fold the
// getelementptr to the resulting integer value casted to the pointer type.
- if (BaseIsInt) {
+ APInt BasePtr(BitWidth, 0);
+ if (ConstantExpr *CE = dyn_cast<ConstantExpr>(Ptr))
+ if (CE->getOpcode() == Instruction::IntToPtr)
+ if (ConstantInt *Base = dyn_cast<ConstantInt>(CE->getOperand(0))) {
+ BasePtr = Base->getValue();
+ BasePtr.zextOrTrunc(BitWidth);
+ }
+ if (Ptr->isNullValue() || BasePtr != 0) {
Constant *C = ConstantInt::get(Ptr->getContext(), Offset+BasePtr);
return ConstantExpr::getIntToPtr(C, ResultTy);
}
// we eliminate over-indexing of the notional static type array bounds.
// This makes it easy to determine if the getelementptr is "inbounds".
// Also, this helps GlobalOpt do SROA on GlobalVariables.
- Ptr = cast<Constant>(Ptr->stripPointerCasts());
const Type *Ty = Ptr->getType();
SmallVector<Constant*, 32> NewIdxs;
do {
if (const SequentialType *ATy = dyn_cast<SequentialType>(Ty)) {
- if (isa<PointerType>(ATy)) {
+ if (ATy->isPointerTy()) {
// The only pointer indexing we'll do is on the first index of the GEP.
if (!NewIdxs.empty())
break;
/// ConstantFoldConstantExpression - Attempt to fold the constant expression
/// using the specified TargetData. If successful, the constant result is
/// result is returned, if not, null is returned.
-Constant *llvm::ConstantFoldConstantExpression(ConstantExpr *CE,
+Constant *llvm::ConstantFoldConstantExpression(const ConstantExpr *CE,
const TargetData *TD) {
SmallVector<Constant*, 8> Ops;
- for (User::op_iterator i = CE->op_begin(), e = CE->op_end(); i != e; ++i) {
+ for (User::const_op_iterator i = CE->op_begin(), e = CE->op_end(); i != e; ++i) {
Constant *NewC = cast<Constant>(*i);
// Recursively fold the ConstantExpr's operands.
if (ConstantExpr *NewCE = dyn_cast<ConstantExpr>(NewC))
case Instruction::ICmp:
case Instruction::FCmp: assert(0 && "Invalid for compares");
case Instruction::Call:
- if (Function *F = dyn_cast<Function>(Ops[0]))
+ if (Function *F = dyn_cast<Function>(Ops[NumOps - 1]))
if (canConstantFoldCallTo(F))
- return ConstantFoldCall(F, Ops+1, NumOps-1);
+ return ConstantFoldCall(F, Ops, NumOps - 1);
return 0;
case Instruction::PtrToInt:
// If the input is a inttoptr, eliminate the pair. This requires knowing
// If the input is a ptrtoint, turn the pair into a ptr to ptr bitcast if
// the int size is >= the ptr size. This requires knowing the width of a
// pointer, so it can't be done in ConstantExpr::getCast.
- if (ConstantExpr *CE = dyn_cast<ConstantExpr>(Ops[0])) {
+ if (ConstantExpr *CE = dyn_cast<ConstantExpr>(Ops[0]))
if (TD &&
- TD->getPointerSizeInBits() <=
- CE->getType()->getScalarSizeInBits()) {
- if (CE->getOpcode() == Instruction::PtrToInt)
- return FoldBitCast(CE->getOperand(0), DestTy, *TD);
-
- // If there's a constant offset added to the integer value before
- // it is casted back to a pointer, see if the expression can be
- // converted into a GEP.
- if (CE->getOpcode() == Instruction::Add)
- if (ConstantInt *L = dyn_cast<ConstantInt>(CE->getOperand(0)))
- if (ConstantExpr *R = dyn_cast<ConstantExpr>(CE->getOperand(1)))
- if (R->getOpcode() == Instruction::PtrToInt)
- if (GlobalVariable *GV =
- dyn_cast<GlobalVariable>(R->getOperand(0))) {
- const PointerType *GVTy = cast<PointerType>(GV->getType());
- if (const ArrayType *AT =
- dyn_cast<ArrayType>(GVTy->getElementType())) {
- const Type *ElTy = AT->getElementType();
- uint64_t AllocSize = TD->getTypeAllocSize(ElTy);
- APInt PSA(L->getValue().getBitWidth(), AllocSize);
- if (ElTy == cast<PointerType>(DestTy)->getElementType() &&
- L->getValue().urem(PSA) == 0) {
- APInt ElemIdx = L->getValue().udiv(PSA);
- if (ElemIdx.ult(APInt(ElemIdx.getBitWidth(),
- AT->getNumElements()))) {
- Constant *Index[] = {
- Constant::getNullValue(CE->getType()),
- ConstantInt::get(ElTy->getContext(), ElemIdx)
- };
- return
- ConstantExpr::getGetElementPtr(GV, &Index[0], 2);
- }
- }
- }
- }
- }
- }
+ TD->getPointerSizeInBits() <= CE->getType()->getScalarSizeInBits() &&
+ CE->getOpcode() == Instruction::PtrToInt)
+ return FoldBitCast(CE->getOperand(0), DestTy, *TD);
+
return ConstantExpr::getCast(Opcode, Ops[0], DestTy);
case Instruction::Trunc:
case Instruction::ZExt:
case Intrinsic::usub_with_overflow:
case Intrinsic::sadd_with_overflow:
case Intrinsic::ssub_with_overflow:
+ case Intrinsic::convert_from_fp16:
+ case Intrinsic::convert_to_fp16:
return true;
default:
return false;
const Type *Ty = F->getReturnType();
if (NumOperands == 1) {
if (ConstantFP *Op = dyn_cast<ConstantFP>(Operands[0])) {
+ if (Name == "llvm.convert.to.fp16") {
+ APFloat Val(Op->getValueAPF());
+
+ bool lost = false;
+ Val.convert(APFloat::IEEEhalf, APFloat::rmNearestTiesToEven, &lost);
+
+ return ConstantInt::get(F->getContext(), Val.bitcastToAPInt());
+ }
+
if (!Ty->isFloatTy() && !Ty->isDoubleTy())
return 0;
/// Currently APFloat versions of these functions do not exist, so we use
return ConstantInt::get(Ty, Op->getValue().countTrailingZeros());
else if (Name.startswith("llvm.ctlz"))
return ConstantInt::get(Ty, Op->getValue().countLeadingZeros());
+ else if (Name == "llvm.convert.from.fp16") {
+ APFloat Val(Op->getValue());
+
+ bool lost = false;
+ APFloat::opStatus status =
+ Val.convert(APFloat::IEEEsingle, APFloat::rmNearestTiesToEven, &lost);
+
+ // Conversion is always precise.
+ status = status;
+ assert(status == APFloat::opOK && !lost &&
+ "Precision lost during fp16 constfolding");
+
+ return ConstantFP::get(F->getContext(), Val);
+ }
return 0;
}
+ if (isa<UndefValue>(Operands[0])) {
+ if (Name.startswith("llvm.bswap"))
+ return Operands[0];
+ return 0;
+ }
+
return 0;
}